CN116757031A - Multi-factor analysis method and device for influencing metal-metal bonding performance - Google Patents
Multi-factor analysis method and device for influencing metal-metal bonding performance Download PDFInfo
- Publication number
- CN116757031A CN116757031A CN202310709558.6A CN202310709558A CN116757031A CN 116757031 A CN116757031 A CN 116757031A CN 202310709558 A CN202310709558 A CN 202310709558A CN 116757031 A CN116757031 A CN 116757031A
- Authority
- CN
- China
- Prior art keywords
- data
- metal
- training
- model
- xgboost algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 229910052751 metal Inorganic materials 0.000 title claims abstract description 85
- 239000002184 metal Substances 0.000 title claims abstract description 85
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000000556 factor analysis Methods 0.000 title claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 134
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 83
- 238000004458 analytical method Methods 0.000 claims abstract description 9
- 238000004088 simulation Methods 0.000 claims description 63
- 238000012360 testing method Methods 0.000 claims description 55
- 238000004381 surface treatment Methods 0.000 claims description 36
- 239000010410 layer Substances 0.000 claims description 34
- 239000003292 glue Substances 0.000 claims description 30
- 229910052782 aluminium Inorganic materials 0.000 claims description 29
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 claims description 29
- 239000012790 adhesive layer Substances 0.000 claims description 19
- 238000002474 experimental method Methods 0.000 claims description 10
- 238000010276 construction Methods 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000002048 anodisation reaction Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 20
- 239000002131 composite material Substances 0.000 abstract description 17
- 239000011248 coating agent Substances 0.000 abstract description 8
- 238000000576 coating method Methods 0.000 abstract description 8
- 238000010801 machine learning Methods 0.000 abstract description 7
- 238000009472 formulation Methods 0.000 abstract description 2
- 239000000203 mixture Substances 0.000 abstract description 2
- 239000007789 gas Substances 0.000 description 15
- NBIIXXVUZAFLBC-UHFFFAOYSA-N Phosphoric acid Chemical compound OP(O)(O)=O NBIIXXVUZAFLBC-UHFFFAOYSA-N 0.000 description 10
- 238000007743 anodising Methods 0.000 description 9
- 239000011247 coating layer Substances 0.000 description 9
- 238000005498 polishing Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 244000137852 Petrea volubilis Species 0.000 description 5
- 229910000147 aluminium phosphate Inorganic materials 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 5
- 239000000463 material Substances 0.000 description 5
- 238000011282 treatment Methods 0.000 description 5
- 238000013461 design Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- MYMOFIZGZYHOMD-UHFFFAOYSA-N Dioxygen Chemical compound O=O MYMOFIZGZYHOMD-UHFFFAOYSA-N 0.000 description 3
- 239000000853 adhesive Substances 0.000 description 3
- 238000004026 adhesive bonding Methods 0.000 description 3
- 230000001070 adhesive effect Effects 0.000 description 3
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000010438 heat treatment Methods 0.000 description 2
- 239000001257 hydrogen Substances 0.000 description 2
- 229910052739 hydrogen Inorganic materials 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 238000013386 optimize process Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 238000005253 cladding Methods 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000007790 scraping Methods 0.000 description 1
- 238000007711 solidification Methods 0.000 description 1
- 230000008023 solidification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/26—Composites
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
Abstract
The invention provides a multi-factor analysis method for influencing metal-metal bonding performance. The analysis method comprises the following steps: acquiring a data set; constructing an Xgboost algorithm model, and defining a hyper-parameter range of the Xgboost algorithm model; performing iterative training for multiple times based on the data set and the Xgboost algorithm model to obtain multiple groups of training results for judging the reliability of the Xgboost algorithm model; determining an optimal model based on a plurality of groups of training results obtained by repeated iterative training; and importing the model interpretable tool SHAP into an optimal model, and outputting interpretation results and prediction results. The analysis method provided by the invention can predict the rule and primary and secondary of the influence of each factor on the metal-metal bonding performance through machine learning, guide the formulation of a bonding process scheme and provide theoretical support for metal coating of the composite gas cylinder.
Description
Technical Field
The invention belongs to the technical field of metal-metal bonding, and particularly relates to a multi-factor analysis method and device for influencing metal-metal bonding performance.
Background
The high-pressure gas cylinders applied to the aerospace technology can be divided into two main types according to application and structure: metal cylinders and composite cylinders. Compared with a metal gas cylinder, the composite gas cylinder has the excellent performances of light weight, low energy consumption, strong bearing capacity, high reliability and the like, and is an ideal pressure container used as a satellite, a space plane, a carrier rocket and a space station for storing various liquids or gases to maintain the normal operation of a subsystem. However, as the service environment of the composite gas cylinder is harsh, the composite gas cylinder is widely applied and faces a plurality of problems, and the problems are mainly expressed in the following two aspects: for a liquid hydrogen cylinder, the ultra-low temperature mechanical property and the permeation of hydrogen molecules need to be overcome, and for a liquid oxygen cylinder, the compatibility problem of materials and liquid oxygen needs to be overcome. Based on the metal coating on the composite material gas cylinder body, the high-pressure gas cylinder with a metal liner-composite material layer-metal coating composite structure is formed, the respective advantages of the metal and the composite material are fully exerted, and the requirements of impact resistance and leakage resistance are met. Wherein, the composite material layer is protected by the double layers of the metal lining and the metal coating layer, so that the problems of compatibility with liquid oxygen and permeability can be effectively avoided.
Considering that the composite gas cylinder is an irregular rotating body, a die is required to be respectively designed according to the profile size of the composite gas cylinder to form a metal coating layer of the bottle mouth, a metal coating layer at the closing-in position of the bottle mouth, a metal coating layer of the bottle bottom and a metal coating layer of the cylinder body through multiple times of stamping, drawing and other processes, and when the composite gas cylinder is subjected to metal coating, in order to enable the metal coating layer to completely coat the composite gas cylinder body, a lap joint process is involved between the metal coating layer of the cylinder body and the metal coating layer of the bottle mouth as well as between the metal coating layer of the bottle bottom, and an experiment of a metal-metal cementing process is required to be carried out.
In the prior art, the process parameter optimization is mainly performed by an orthogonal experiment method, the orthogonal experiment method can only perform the process parameter optimization based on a small amount of data samples, and the optimized process parameter is possibly not optimal due to the fact that the data amount samples are fewer; and for the samples with more data volume, the orthogonal test can not make reasonable experimental design and process parameter optimization, and meanwhile, the method has the defects of longer time consumption, higher consumable cost and the like in the process test.
Disclosure of Invention
The invention aims to provide an analysis method for predicting the importance of multifactor affecting the metal-metal cementing performance based on a machine learning technology, which has the advantages of low experimental cost and high prediction accuracy.
In order to achieve the above object, the present invention provides a multi-factor analysis method for influencing metal-metal bonding performance, comprising the steps of:
(1) Acquiring a data set, wherein the data set comprises a plurality of data packets, each data packet comprises X data consisting of a level of multiple factors affecting metal-metal bonding performance and Y data consisting of a pull shear strength true value, and the X data and the Y data in each data packet have a one-to-one correspondence;
(2) Constructing an Xgboost algorithm model, and defining a hyper-parameter range of the Xgboost algorithm model;
(3) Performing iterative training for multiple times based on the data set and the Xgboost algorithm model to obtain multiple groups of training results for judging the reliability of the Xgboost algorithm model, wherein the training results comprise root mean square errors and regression coefficients, and the hyper-parameters of the Xgboost algorithm model corresponding to each iterative training are optimized and updated in the hyper-parameter range of the set Xgboost algorithm model; wherein,,
each iterative training includes: randomly dividing the data set into a training set and a testing set according to a preset proportion, and training the Xgboost algorithm model by taking the training set as input data to obtain a trained Xgboost algorithm model; inputting the test set as input data into a trained Xgboost algorithm model, and acquiring a tensile and shear strength predicted value corresponding to the X data of each data packet in the test set based on the X data of all the data packets in the test set; based on the pull-shear strength real value and the pull-shear strength predicted value corresponding to the X data of each data packet in the test set, obtaining a training result corresponding to the iterative training, wherein the pull-shear strength real value is Y data corresponding to the X data;
(4) Based on a plurality of groups of training results obtained by repeated iterative training, determining an optimal model, wherein the root mean square error in the training results corresponding to the optimal model is the smallest, and the regression coefficient is larger than or equal to a preset regression coefficient;
(5) And importing a model interpretable tool SHAP into the optimal model, and outputting interpretation results and prediction results, wherein the interpretation results comprise influence values of multiple factors affecting the metal-metal bonding performance on tensile shear strength respectively.
In a specific embodiment, the multifactor includes a lap length, a glue layer thickness and a surface treatment mode, and the interpretation result further includes an importance ranking of the influence of the multifactor on the tensile shear strength, where the importance ranking includes the surface treatment mode, the glue layer thickness and the lap length in sequence from high to low.
In a specific embodiment, the predicted results include lap length, bond line thickness, and surface treatment pattern corresponding to optimal metal-to-metal bonding performance.
In a specific embodiment, the overlap length corresponding to the best metal-to-metal bond performance is 26mm, the bond line thickness is 0.1mm, and the surface treatment is 800# sanding + anodization.
In a specific embodiment, the step of acquiring the data set comprises:
The method comprises the steps of establishing a geometric model of metal lap joint by adopting Abaqus software, wherein the geometric model consists of two layers of aluminum plates and a glue layer clamped between the two aluminum plates, and one layer of aluminum plate is partially lapped on the other layer of aluminum plate;
setting cohesive force contact between the aluminum plate of the geometric model and the adhesive layer, and carrying out grid division and boundary condition setting to construct a finite element model for metal-metal bonding;
verifying the accuracy of the finite element model to obtain a target finite element model with an error smaller than a preset error;
determining N simulation schemes according to multiple factors and multiple levels affecting metal-metal bonding performance, wherein the factor levels comprise overlap joint length, adhesive layer thickness and surface treatment mode, and N is an integer of more than or equal to 500 and less than or equal to 1000;
taking the N simulation schemes as input, and performing simulation by using the target finite element model to obtain the maximum tensile shear load corresponding to each simulation scheme;
based on the maximum tensile shear load corresponding to each simulation scheme, acquiring tensile shear strength corresponding to each simulation scheme;
and combining each simulation scheme with the pull-shear strength corresponding to each simulation scheme to obtain N simulation data, wherein each simulation data corresponds to one data packet, the simulation scheme is X data, and the pull-shear strength is Y data.
In a specific embodiment, the step of verifying the accuracy of the finite element model includes:
step A, establishing an orthogonal table according to multi-factor multi-level affecting metal-metal bonding performance, wherein the multi-factor comprises overlap joint length, adhesive layer thickness and surface treatment mode;
step B, performing metal-metal bonding experiments according to M bonding schemes determined by the orthogonal table, and obtaining a maximum experimental tensile shear load corresponding to each bonding scheme, wherein M is an integer which is more than 8 and less than 15;
step C, taking the M cementing schemes as input, and performing simulation by using the finite element model to obtain the maximum simulation tensile shear load corresponding to each cementing scheme;
step D, calculating an error based on the maximum experimental pull shear load and the maximum simulation pull shear load corresponding to each cementing scheme;
and E, comparing the absolute value of the error with a preset error, and if the absolute value of the error is smaller than the preset error, taking the finite element model as a target finite element model.
In a specific embodiment, the step of verifying the accuracy of the finite element model further comprises:
and (C) when the absolute value of the error is larger than a preset error, updating the input parameters including the initial rigidity, critical traction and fracture energy of the corresponding adhesive layer when the cohesive force contact is set to obtain a new finite element model of metal-metal bonding, and repeating the steps (C) to (E).
In a specific embodiment, the step (4) includes:
establishing a one-to-one correspondence between root mean square errors and regression coefficients in each group of training results obtained based on multiple iterative training;
sequencing the root mean square errors in all the training results, and determining the minimum root mean square error;
and when the regression coefficient corresponding to the minimum root mean square error is larger than or equal to a preset regression coefficient, determining the Xgboost algorithm model corresponding to the minimum root mean square error as an optimal model.
In a specific embodiment, the iteration termination condition of the multiple iteration training in the step (3) is that the iteration number is 80-150.
The present invention also provides an analysis apparatus for multi-factors affecting metal-to-metal bonding performance, the analysis apparatus comprising:
the acquisition module is used for acquiring a data set, wherein the data set comprises a plurality of data packets, each data packet comprises X data consisting of a multi-factor level affecting metal-metal bonding performance and Y data consisting of a tensile shear strength true value, and the X data and the Y data in each data packet have a one-to-one correspondence;
the construction module is used for constructing an Xgboost algorithm model and defining a hyper-parameter range of the Xgboost algorithm model;
The training module is used for carrying out iterative training for a plurality of times based on the data set and the Xgboost algorithm model to obtain a plurality of groups of training results for judging the reliability of the Xgboost algorithm model, wherein the training results comprise root mean square errors and regression coefficients, and the hyper-parameters of the Xgboost algorithm model corresponding to each iterative training are optimally updated in the hyper-parameter range of the set Xgboost algorithm model; wherein,,
each iterative training includes: cutting the data set into a training set and a testing set according to a preset proportion, and training the Xgboost algorithm model by taking the training set as input data to obtain a trained Xgboost algorithm model; inputting the test set as input data into a trained Xgboost algorithm model, and acquiring a tensile and shear strength predicted value corresponding to the X data of each data packet in the test set based on the X data of all the data packets in the test set; based on the pull-shear strength real value and the pull-shear strength preset value corresponding to the X data of each data packet in the test set, acquiring a training result corresponding to the iterative training, wherein the pull-shear strength real value is Y data corresponding to the X data;
the determining module is used for determining an optimal model based on a plurality of groups of training results obtained by repeated iterative training, wherein the root mean square error in the training results corresponding to the optimal model is minimum, and the regression coefficient is larger than or equal to a preset regression coefficient;
And the output module is used for leading a model interpretable tool SHAP into the optimal model and outputting interpretation results and prediction results, wherein the interpretation results comprise influence values of multiple factors affecting the metal-metal bonding performance on tensile shear strength respectively.
The beneficial effects of the invention at least comprise:
the invention provides a multi-factor analysis method for influencing metal-metal bonding performance, which comprises the steps of training a model through a training set which is randomly segmented and an Xgboost algorithm model which is optimized and updated each time of iteration, calculating and judging the training result of the reliability of the Xgboost algorithm model through a test set, obtaining a group of training results through each time of iteration training, comparing all training results after the iteration times reach preset times, determining the model corresponding to the best training result as an optimal model, then importing a model interpretable tool SHAP into the optimal model, and outputting an interpretation result and a prediction result; therefore, the law and the primary and secondary of the influence of each factor on the metal-metal bonding performance can be predicted through machine learning, the formulation of a bonding process scheme is guided, and theoretical support is provided for the metal coating of the composite gas cylinder; compared with the orthogonal process test, the number of tests and the consumable cost are greatly reduced, and the process parameters of the optimizing process are more accurate.
In addition to the objects, features and advantages described above, the present invention has other objects, features and advantages. The present invention will be described in further detail with reference to the drawings.
Drawings
FIG. 1 is a flow chart illustrating a method for multi-factor analysis of metal-to-metal bonding performance according to an embodiment of the present invention;
FIG. 2 is a block diagram of a finite element model constructed in accordance with an embodiment of the present invention;
FIG. 3 is a graph comparing the maximum experimental pull shear load and the maximum simulated pull shear load corresponding to 10 bonding schemes according to an embodiment of the present invention;
FIG. 4 is a graph comparing the true values of tensile shear strength with the predicted values of tensile shear strength provided by the invention;
FIG. 5 is a ranking chart of importance of each factor of the output of the optimal model provided by the invention;
FIG. 6 is a graph of optimal process parameters for optimal model prediction provided by the present invention;
FIG. 7 is a block diagram of a multi-factor analysis method for influencing metal-to-metal bonding performance according to one embodiment of the present invention.
Detailed Description
The embodiments of the invention are described in detail below with reference to the attached drawings, but the invention can be defined and covered in a number of different embodiments according to the claims.
Orthogonal experimentation is a common method for performing process experimental optimization, but in contrast, it can only perform process parameter optimization based on a small number of data samples, and because of the small number of data samples, the optimized process parameters may not be optimal. And for the samples with more data volume, the orthogonal test can not reasonably design experiments and optimize technological parameters, and meanwhile, the method has the defects of long experiment time consumption and high consumable cost.
The invention can make up for the deficiency by combining the finite element method in order to obtain more sample data. With the explosive development of artificial intelligence and big data technology, machine learning is a very promising technology, and has more and more obvious advantages in terms of data model fitting and data mining, can utilize a large amount of accumulated experimental data to improve new material design and research and development efficiency, greatly reduce time and cost consumption, and has been successfully applied to the fields of material science, engineering manufacture and the like, meanwhile, the machine learning technology is a powerful tool for revealing and describing complex relations between material performance and various features, and can visually display the complex relations between various features and target output.
The Xgboost algorithm model is used as one of machine learning Boosting models, and a strong classifier is formed by integrating a plurality of tree models, so that the method has the advantages that the prediction capability of a prediction model can be improved, a regular term is added in a cost function for controlling the complexity of the model, the variance of the model is reduced, the learned model is simpler, and the overfitting risk of the model is reduced.
According to the invention, the data set is acquired through finite element simulation, so that the test times and the consumable cost are greatly reduced, the metal-metal cementing performance is predicted and the importance ranking of the model output (cementing tensile shear strength) is evaluated based on the machine learning technology (lap joint length, glue layer thickness and surface treatment mode), and a set of more reasonable process parameters are optimized, so that theoretical support is provided for metal cladding of the composite material gas cylinder.
Referring to fig. 1 in combination, the present invention provides a multi-factor analysis method for influencing metal-metal bonding performance, the analysis method comprising the following steps:
step S10, acquiring a data set, wherein the data set comprises a plurality of data packets, each data packet comprises X data consisting of a multi-factor level affecting metal-metal bonding performance and Y data consisting of a pull shear strength true value, and the X data and the Y data in each data packet have a one-to-one correspondence;
in this embodiment, the multiple factors include overlap length, glue line thickness, and surface treatment.
Each data packet comprises X data consisting of the level of the lap length, the level of the glue layer thickness and the level of the surface treatment mode and Y data consisting of the true value of the tensile shear strength, wherein the true value of the tensile shear strength corresponds to the level of the lap length, the level of the glue layer thickness and the level of the surface treatment mode. The X data can be understood as variables and the Y data as a result of the determination by the variables.
In this embodiment, the data set is composed of simulation data obtained based on finite element simulation, and the true value of tensile shear strength is a simulation result obtained by using the finite element simulation with X data as input data.
In this embodiment, the data set may include other amounts of data such as 500 sets of data, 600 sets of data, 800 sets of data, etc., which is not limited herein, and it is understood that the larger the data amount of the data set is, the larger the corresponding amount of computation will be, and the more accurate the prediction result will be.
Preferably, the step of acquiring the data set comprises:
the method comprises the following steps of (1) establishing a geometric model of metal lap joint by adopting Abaqus software, wherein the geometric model consists of two layers of aluminum plates and a glue layer clamped between the two aluminum plates, and one layer of aluminum plate is partially lapped on the other layer of aluminum plate;
step (2) cohesive force contact is arranged between the aluminum plate of the geometric model and the adhesive layer, grid division and boundary condition setting are carried out, and a finite element model of metal-metal bonding is constructed;
parameters setting cohesive force contact inputs include initial stiffness, critical traction, and fracture energy of the corresponding bond line.
The grid division and boundary condition setting are specifically as follows: referring to fig. 2 in combination, the overall mesh size of the aluminum plate and the glue layer is 1, wherein the total number of aluminum plate units is 6250, the mesh type is C3D8R linear hexahedral units, the total number of glue layer units is determined according to the overlap length of the glue layer bonding area (when the glue layer length is 15mm and the width is 25mm, the total number of corresponding glue layer units is 375), the mesh type COH3D8 linear hexahedral units are divided into meshes in a sweep mode in the thickness direction. The reference point RP-2 is coupled with the fixed end, the boundary condition is ENCASTRE, the reference point RP-1 is coupled with the loading end, and displacement constraint is applied.
Step (3) verifying the accuracy of the finite element model to obtain a target finite element model with an error smaller than a preset error;
by verifying the accuracy of the established finite element model, the accuracy of the data of the training set and the testing set can be ensured, so that the prediction accuracy is improved.
Preferably, the step of verifying the accuracy of the finite element model comprises:
step A, establishing an orthogonal table according to multi-factor multi-level affecting metal-metal bonding performance, wherein the multi-factor comprises overlap joint length, adhesive layer thickness and surface treatment mode;
in this embodiment, the multiple factors include a lap length, a glue line thickness, and a surface treatment mode, wherein the levels corresponding to the lap length include 15mm, 20mm, and 25mm, the levels corresponding to the glue line thickness include 0.1 mm, 0.2mm, and 0.3 mm, and the levels corresponding to the surface treatment mode include 800# sanding+anodizing, 400# sanding+anodizing, 1200# sanding+anodizing.
Based on the predetermined multi-factor and multi-factor levels, the test design was performed using an orthogonal test design method, and an orthogonal table was established as shown in table 1.
TABLE 1 orthogonal tables for multifactor and multilevel correspondence
Remarks: the values in the surface treatment mode represent the mesh number polishing of the coated abrasive + the anodizing treatment.
As can be seen from Table 1, the orthogonal tables are co-paired10 cementing schemes are adopted, namely a first cementing scheme corresponding to the serial number 1, the lap joint length is 25mm, the glue layer thickness is 0.3mm and the surface treatment mode is 800#Polishing by sand paper; a second cementing scheme corresponding to the serial number 2, a lap joint length of 20mm, a glue layer thickness of 0.1mm and a surface treatment mode of 1200#Polishing by sand paper; a third cementing scheme corresponding to the serial number 3, a lap joint length of 15mm, a glue layer thickness of 0.1mm and a surface treatment mode of 800#Polishing by sand paper; … … corresponds to a gluing scheme, and will not be described in detail.
Step B, performing metal-metal bonding experiments according to M bonding schemes determined by the orthogonal table, and obtaining a maximum experimental tensile shear load corresponding to each bonding scheme, wherein M is an integer which is more than 8 and less than 15;
for easy understanding, the metal-metal bonding test performed by way of example for the bonding scheme corresponding to the number 1 of the orthogonal table is described as follows:
providing a first aluminum plate and a second aluminum plate, and respectively preprocessing the first aluminum plate and the second aluminum plate, wherein the preprocessing comprises sanding treatment and phosphoric acid anodizing treatment which are sequentially carried out on a lapping area of the first metal plate and a lapping area of the second metal plate, and the mesh number of the sandpaper is 800;
The overlapping area of the first aluminum plate and the overlapping area of the second aluminum plate refer to the area where the first aluminum plate and the second aluminum plate are connected.
Coating DW-3 low-temperature glue with preset thickness on the lapping area of the first aluminum plate and the lapping area of the first aluminum plate by using a glue scraping tool, wherein the preset thickness is 0.15mm, and the length (lapping length) of the lapping area is 25mm;
connecting the first aluminum plate and the second aluminum plate through DW-3 low-temperature glue to obtain a lap joint structure to be solidified, and then placing the lap joint structure to be solidified in an autoclave for solidification to obtain a solidified lap joint structure;
the curing process can be as follows: heating to 60 ℃ according to a heating rate of 1.5 ℃/min, preserving heat at 60 ℃ for 480min, cooling along with a furnace, vacuumizing to 0.1MPa in the vacuum bag in the whole curing process, and pressurizing to 0.05MPa in the autoclave.
Measuring the maximum tensile shear load of the lap joint structure, wherein the obtained maximum tensile shear load is 10717.1N;
calculating the tensile shear strength according to a preset formula, wherein the obtained tensile shear strength is 17.1/MPa, and the preset formula is as follows:
τ is the tensile shear strength of the adhesive and MPa; p is the maximum load required by pull shear damage and N; b and L are the width and length of the sample overlap region, respectively, and mm.
All the steps of the cementing test are the same, and the difference is only that the lap joint length is different or the glue layer thickness is different or the number of sand paper meshes is different during polishing treatment.
The respective bonding schemes and their corresponding experimental results are shown in table 2:
table 2 group 10 cement bonding scheme and mechanical properties thereof
Remarks: the values in the surface treatment mode represent the mesh number polishing of the coated abrasive + the anodizing treatment.
Step C, taking the M cementing schemes as input, and performing simulation by using the finite element model to obtain the maximum simulation tensile shear load corresponding to each cementing scheme;
step D, calculating an error based on the maximum experimental pull shear load and the maximum simulation pull shear load corresponding to each cementing scheme;
and E, comparing the error with a preset error, and if the error is smaller than the preset error, taking the finite element model as a target finite element model.
For convenience of explanation, the maximum simulated tensile shear load corresponding to the cementing scheme is marked as P si The maximum to be corresponding to the cementing schemeThe load of the experimental tensile shear is marked as P vi The calculation formula of the error E is:
E＝(P vi -P si )/P vi *100％
and (4) when the absolute value of the error is smaller than a preset error, the accuracy of the finite element model is high, the finite element model is a target finite element model, and the target finite element model is used for simulation in the step (4) so as to obtain the maximum tensile shear load corresponding to N simulation schemes one by one.
When the absolute value of the error is larger than a preset error, updating the parameters input when cohesive force contact is set to obtain a new finite element model of metal-metal bonding, and verifying the accuracy of the new finite element model of metal-metal bonding, wherein the method comprises the steps of simulating by using the new finite element model of metal-metal bonding, re-obtaining the maximum simulated tensile shear load corresponding to 10 bonding schemes, re-calculating the error and the like, namely repeating the steps C to E again; and until the absolute value of the error is smaller than a preset error, wherein the input parameters comprise initial rigidity, critical traction force and fracture energy of the corresponding adhesive layer.
In this embodiment, the preset error is 6%, if the calculated error E is between [ 6%6% ], the absolute value of the error is smaller than the preset error, and the pull-shear strength of each simulation scheme calculated by using the finite element model may be calculated.
In this example, the initial stiffness, the critical traction force, and the fracture energy are shown in tables 3 and 4.
TABLE 3 mechanical Properties of adhesive layer normal tension
TABLE 4 mechanical Properties of adhesive layer shear test under different surface treatments
Referring to fig. 3 in combination, fig. 3 is a comparison chart of the maximum experimental pull shear load and the maximum simulated pull shear load corresponding to 10 gluing schemes, and as can be seen from fig. 3, the difference between the simulated data and the experimental data is smaller, and the error is within the preset error range.
Step (4) determining N simulation schemes according to multiple factors and multiple levels affecting metal-metal bonding performance, wherein the factor levels comprise overlap joint length, adhesive layer thickness and surface treatment mode, and N is an integer of more than or equal to 500 and less than or equal to 1000;
in this embodiment, the multiple factors include a lap length, a glue layer thickness and a surface treatment mode, wherein the value range of the level corresponding to the lap length of the factors is 10-30 mm, and the increment is 1mm, that is, the level corresponding to the lap length includes 10mm, 11mm, 12mm, 13mm … … mm, 29mm and 30mm, which are not exhaustive, that is, the initial value is 10mm, and then the next lap length is determined according to the increment of 1 mm; the value range of the level corresponding to the thickness of the factor adhesive layer is 0.1-1 mm, and the increment is 0.1mm, namely the level corresponding to the thickness of the adhesive layer comprises 0.1mm, 0.2mm, 0.3mm, 0.4mm, 0.5mm, 0.6mm, 0.7mm, 0.8mm, 0.9mm and 1.0mm; the surface treatment mode is one of 400# + phosphoric acid anodized, 800# + phosphoric acid anodized and 1200# + phosphoric acid anodized.
And combining the lap length level, the glue layer thickness level and the surface treatment mode level to obtain N simulation schemes, wherein each simulation scheme comprises the lap length level, the glue layer thickness level and the surface treatment mode level.
Step (5) taking the N simulation schemes as input, and performing simulation by using the target finite element model to obtain the maximum tensile shear load corresponding to each simulation scheme;
step (6), based on the maximum tensile shear load corresponding to each simulation scheme, acquiring the tensile shear strength corresponding to each simulation scheme;
based on the maximum tensile shear load, the tensile shear strength is calculated according to a preset formula, wherein the preset formula is as follows:
τ is the tensile shear strength of the adhesive and MPa; p is the maximum load required by pull shear damage and N; b and L are the width and length of the sample overlap region, respectively, and mm.
And (7) combining each simulation scheme with pull-shear strength corresponding to each simulation scheme to obtain N simulation data, wherein each simulation data corresponds to one data packet, the simulation scheme is X data, and the pull-shear strength is Y data.
In this embodiment, the number of simulation data is 630, i.e. the data set comprises 630 data packets. It can be understood that each simulation data is a data packet, the simulation scheme corresponding to each simulation data is the X data in the data packet, the pull-shear strength corresponding to each simulation data is the Y data in the data packet, that is, the pull-shear strength obtained by simulation is the pull-shear strength true value in the data packet.
In the invention, the accuracy of the finite element simulation is judged by cementing experimental data, and the finite element simulation with high accuracy is utilized to acquire the data set, so that on one hand, the accuracy of the simulation data is high, the prediction accuracy can be effectively improved, and on the other hand, the cost of the manual experiment and the error of the manual experiment can be reduced by acquiring the data set based on the finite element simulation.
In this embodiment, N sets of data obtained by finite element simulation are integrated into an Excel file, and all sets of data are read in Jupyter Notebook.
S20, constructing an Xgboost algorithm model, and defining a hyper-parameter range of the Xgboost algorithm model;
building an Xgboost algorithm model, importing the Xgboost algorithm model, setting a hyper-parameter range of the model, and specifically: boost= [ gbtree "," dart ", eta= [ 0.1,0.2,0.3,0.4,0.5 ], gamma= [ 0,1 ], max_depth= [ 2,4,6,8 ], min_child_weight= [ 0,1 ], reg_lamda= [ 0.1,0.5,1,2 ], objective=" reg: squarederror ".
In this embodiment, the initial value of the Xgboost algorithm model hyper-parameter is: boost= "gbtree", eta=0.3, gamma=0, max_depth=6, min_child_weight=1, reg_lamda=1, objective= "reg: squarederror", wherein: boost= 'gbtree' represents each iteration run using a tree-based model; eta is learning rate, default value is 0.3, step size shrinkage is used in updating to prevent overfitting; gamma is the minimum value of loss function reduction required by specified splitting, the default value is 0, when a node is split, the loss function reduction value is split only when the gamma node is greater than or equal to the gamma node, the algorithm is more conservative when the gamma value is greater, fitting is not easy, the performance is not necessarily guaranteed, and balance is required; max_depth represents the maximum depth of a tree, the default value is 6, and increasing this value will make the model more complex, resulting in an overfitting; min_child_weight is used to control the over-fit, and the default value is 1, which if too high, can easily result in under-fit. reg_lambda represents an L2 regularized weight term, the default value is 1, and increasing this value will make the model more conservative; the objective= "reg: squarederror" means that a least squares error loss function is defined.
Step S30, performing multiple iterative training based on the data set and the Xgboost algorithm model to obtain multiple groups of training results for judging the reliability of the Xgboost algorithm model, wherein the training results comprise root mean square errors and regression coefficients, and the hyper-parameters of the Xgboost algorithm model corresponding to each iterative training are optimally updated in the hyper-parameter range of the set Xgboost algorithm model; wherein,,
each iterative training includes: randomly dividing the data set into a training set and a testing set according to a preset proportion, and training the Xgboost algorithm model by taking the training set as input data to obtain a trained Xgboost algorithm model; inputting the test set as input data into a trained Xgboost algorithm model, and acquiring a tensile and shear strength predicted value corresponding to the X data of each data packet in the test set based on the X data of all the data packets in the test set; based on the pull-shear strength real value and the pull-shear strength predicted value corresponding to the X data of each data packet in the test set, obtaining a training result corresponding to the iterative training, wherein the pull-shear strength real value is Y data corresponding to the X data;
it can be understood that the calculation training result is calculated based on the Y data corresponding to the X data in the test set and the pull shear strength prediction value corresponding to the X data predicted by the Xgboost algorithm model.
According to the invention, the Xgboost algorithm model is trained by taking the minimum root mean square error as a target through the combination of random acquisition of the training set and the hyper-parameters of the Xgboost algorithm model, and a group of training results are acquired each time the Xgboost algorithm model is tested through the testing set.
In this embodiment, the preset ratio is 4:1, i.e. the training set is 80% of the data set and the test set is 20% of the data set.
Each time training is iterated, the data set is randomly segmented again to obtain different training sets and test sets.
In this embodiment, the training set includes 630×0.8=504 packets, and the test set includes 630×0.2=128 packets.
Preferably, the number of iterative training is 80 to 150, and in this embodiment, the number of iterative training is 100.
For easy understanding, the first iterative training process includes randomly dividing a training set composed of 630 data packets to obtain a first iterative training set and a first iterative testing set, training an initial Xgboost algorithm model by using the first iterative training set to obtain a first iterative trained Xgboost algorithm model, and then calculating tensile shear strength predicted values corresponding to X data in 128 data packets in the testing set by using the first iterative training Xgboost algorithm model as input, and calculating according to a calculation formula of mean square error and a formula of regression coefficient to obtain training results comprising root mean square error and regression coefficient respectively based on Y data (tensile shear strength true values) corresponding to the X data and tensile shear strength predicted values corresponding to the X data; because the iteration times are less than 100 times, the condition of iteration termination is not satisfied, and the second iteration training is continued; based on the set hyper-parameter range, adjusting the value of the hyper-parameter to obtain a second Xgboost algorithm model corresponding to the second iteration, randomly dividing a training set consisting of 630 data packets to obtain a training set of the second iteration and a testing set of the second iteration, training the second Xgboost algorithm model by using the training set of the second iteration to obtain an Xgboost algorithm model trained by the second iteration, and then taking the testing set of the second iteration as input, calculating pull-shear strength predicted values corresponding to X data in 128 data packets in the testing set by using the Xgboost algorithm model trained by the second iteration, and respectively calculating according to a calculation formula of mean square error and a formula of regression coefficient based on Y data (pull-shear strength real values) corresponding to the X data and the pull-shear strength predicted values corresponding to the X data to obtain training results comprising root mean square error and regression coefficient; because the iteration times are less than 100 times, the condition of iteration termination is not satisfied, and the third iteration training is continued; … …, which is not exhaustive, differs from each iteration of training primarily in that the Xgboost algorithm model is different, and the training set and the test set are different. After 100 iterative training, 100 sets of training results are obtained, wherein each set of training results comprises root mean square error and regression coefficient, and the root mean square error and the regression coefficient are corresponding to each other.
The Root Mean Square Error (RMSE) is calculated as:
wherein N is the total number of samples, y ai To the true value of the tensile and shear strength, y pi Predicted value of tensile shear strength
Regression coefficient (R) 2 ) The calculation formula of (2) is as follows:
wherein N is the total number of samples, y ai To the true value of the tensile and shear strength, y pi In order to predict the tensile and shear strength values,is the average of all the values of the true values of the tensile shear strength.
Specifically, the true tensile shear strength value is derived from a test set, and the predicted tensile shear strength value is calculated through an Xgboost algorithm model.
Step S40, determining an optimal model based on a plurality of groups of training results obtained by repeated iterative training, wherein the root mean square error in the training results corresponding to the optimal model is minimum, and the regression coefficient is greater than or equal to a preset regression coefficient;
the method comprises the following steps:
establishing a one-to-one correspondence between root mean square errors and regression coefficients in each group of training results obtained based on multiple iterative training;
sequencing the root mean square errors in all the training results, and determining the minimum root mean square error;
and when the regression coefficient corresponding to the minimum root mean square error is larger than or equal to a preset regression coefficient, determining the Xgboost algorithm model corresponding to the minimum root mean square error as an optimal model.
Normally, the root mean square error is minimum, and the corresponding regression coefficient is also maximum, so that the optimal model can be determined through the root mean square error; the regression coefficient is added for further judgment in order to screen out abnormal data.
In this example, the preset regression coefficient is 0.95.
The Xgboost algorithm model corresponding to the minimum root mean square error can be understood as: if the root mean square error obtained by the 82 nd iteration training is the minimum root mean square error, the Xgboost algorithm model corresponding to the 82 th iteration training is the optimal model.
In this embodiment, the super parameters corresponding to the optimal model are: boost= 'gbtree', eta=0.2, gamma=0, max_depth=2, min_child_weight=0, reg_lambda=0.2, and objective= "reg: squarederror".
Referring to fig. 4 in combination, the identity in fig. 4 indicates that the experimentally measured tensile shear strength value is exactly equal to the predicted tensile shear strength value, and is used as a reference line for displaying and referencing the regression lineThe lines have a certain deviation, wherein the regression coefficient R is used for 2 A model prediction reliability of 0.995 is better.
And S50, importing a model interpretable tool SHAP into an optimal model, and outputting interpretation results and prediction results, wherein the interpretation results comprise influence values of multiple factors affecting metal-metal adhesive joint performance on tensile shear strength respectively.
The imported model may interpret the tool SHAP (Shapley Additive exPlanations) to evaluate the impact of various features on tensile shear strength. Referring to fig. 5 in combination, it can be seen from fig. 5 that the influence value of the surface treatment mode is +0.55, the influence value of the lap length is +0.24, and the influence value of the glue layer thickness is +0.26; the importance order of the influence of the multifactor on the metal-metal adhesive bonding performance is sequentially from high to low, namely a surface treatment mode, adhesive layer thickness and overlap joint length.
It should be noted that "+" represents a positive effect and "-" represents a negative effect.
Preferably, the predicted result includes overlap length, bond line thickness and surface treatment pattern corresponding to the optimal metal-to-metal bonding performance. Referring to FIG. 6 in combination, it can be seen from FIG. 6 that the average tensile shear strength predicted by the model output is 15.82MPa, and the tensile shear strength predicted by the model is 16.27M when the lap length is 26mm, the bond thickness is 0.1mm, and the surface treatment mode is 800# sanding+phosphoric acid anodizingPa, has the best metal-metal bonding performance. The optimal process parameters can be determined by lapping 26mm in length and 0.1mm in thickness of the adhesive layer, and the surface treatment mode is 800# sanding and anodizing.
When the above prediction results are applied to metal coating of a composite gas cylinder, the values of various parameters cannot be controlled very accurately, and certain errors are definitely present, and in combination with actual operation, the proposal is that: the lap joint length value is controlled within 26mm plus or minus 2mm, the thickness of the adhesive layer is as thin as possible and is close to 0.1mm, and the surface treatment mode of the metal plate material of the coating is 800# sand paper polishing and anodizing.
The present invention also provides an analysis apparatus for multi-factors affecting metal-to-metal bonding performance, the analysis apparatus 100 comprising:
an acquisition module 10, configured to acquire a data set, where the data set includes a plurality of data packets, each of the data packets includes X data including a level of multiple factors affecting metal-metal bonding performance and Y data including a true value of tensile shear strength, and the X data and the Y data in each of the data packets have a one-to-one correspondence;
the construction module 20 is used for constructing an Xgboost algorithm model and defining the hyper-parameter range of the Xgboost algorithm model;
the training module 30 is configured to perform multiple iterative training based on the data set and the Xgboost algorithm model, obtain multiple sets of training results for determining reliability of the Xgboost algorithm model, where the training results include root mean square error and regression coefficients, and perform optimization update on the hyper-parameters of the Xgboost algorithm model corresponding to each iterative training within the set hyper-parameters range of the Xgboost algorithm model; wherein,,
Each iterative training includes: cutting the data set into a training set and a testing set according to a preset proportion, and training the Xgboost algorithm model by taking the training set as input data to obtain a trained Xgboost algorithm model; inputting the test set as input data into a trained Xgboost algorithm model, and acquiring a tensile and shear strength predicted value corresponding to the X data of each data packet in the test set based on the X data of all the data packets in the test set; based on the pull-shear strength real value and the pull-shear strength predicted value corresponding to the X data of each data packet in the test set, obtaining a training result corresponding to the iterative training, wherein the pull-shear strength real value is Y data corresponding to the X data;
the determining module 40 is configured to determine an optimal model based on a plurality of sets of training results obtained by multiple iterative training, where a root mean square error in the training results corresponding to the optimal model is the smallest and the regression coefficient is greater than or equal to a preset regression coefficient;
and an output module 50 for importing a model interpretable tool SHAP into the optimal model and outputting interpretation results and prediction results, wherein the interpretation results comprise influence values of multiple factors affecting the metal-metal bonding performance on tensile shear strength respectively.
It will be clear to those skilled in the art that, for convenience and brevity of description, reference may be made to the details of the foregoing method embodiments for the specific operation of the apparatus and modules described above, which are not described in detail herein.
The foregoing is a further detailed description of the invention in connection with specific preferred embodiments, and is not intended to limit the practice of the invention to such description. It will be apparent to those skilled in the art that several simple deductions and substitutions can be made without departing from the spirit of the invention, and these are considered to be within the scope of the invention.
Claims (10)
1. A method of multi-factor analysis affecting metal-to-metal bonding performance comprising the steps of:
(1) Acquiring a data set, wherein the data set comprises a plurality of data packets, each data packet comprises X data consisting of a level of multiple factors affecting metal-metal bonding performance and Y data consisting of a pull shear strength true value, and the X data and the Y data in each data packet have a one-to-one correspondence;
(2) Constructing an Xgboost algorithm model, and defining a hyper-parameter range of the Xgboost algorithm model;
(3) Performing iterative training for multiple times based on the data set and the Xgboost algorithm model to obtain multiple groups of training results for judging the reliability of the Xgboost algorithm model, wherein the training results comprise root mean square errors and regression coefficients, and the hyper-parameters of the Xgboost algorithm model corresponding to each iterative training are optimized and updated in the hyper-parameter range of the set Xgboost algorithm model; wherein,,
each iterative training includes: randomly dividing the data set into a training set and a testing set according to a preset proportion, and training the Xgboost algorithm model by taking the training set as input data to obtain a trained Xgboost algorithm model; inputting the test set as input data into a trained Xgboost algorithm model, and acquiring a tensile and shear strength predicted value corresponding to the X data of each data packet in the test set based on the X data of all the data packets in the test set; based on the pull-shear strength real value and the pull-shear strength predicted value corresponding to the X data of each data packet in the test set, obtaining a training result corresponding to the iterative training, wherein the pull-shear strength real value is Y data corresponding to the X data;
(4) Based on a plurality of groups of training results obtained by repeated iterative training, determining an optimal model, wherein the root mean square error in the training results corresponding to the optimal model is the smallest, and the regression coefficient is larger than or equal to a preset regression coefficient;
(5) And importing a model interpretable tool SHAP into the optimal model, and outputting interpretation results and prediction results, wherein the interpretation results comprise influence values of multiple factors affecting the metal-metal bonding performance on tensile shear strength respectively.
2. The method according to claim 1, wherein the multifactor includes a lap length, a glue layer thickness, and a surface treatment mode, and the interpretation result further includes an importance ranking of the influence of multifactor on the tensile shear strength, the importance ranking being the surface treatment mode, the glue layer thickness, and the lap length in order from high to low.
3. The method according to claim 2, wherein the predicted result includes a lap length, a glue line thickness, and a surface treatment pattern corresponding to the optimal metal-metal bonding performance.
4. A multi-factor analysis method for influencing metal-metal bond performance according to claim 3 wherein the lap length corresponding to the best metal-metal bond performance is 26mm, the bond line thickness is 0.1mm, and the surface treatment is 800# sanding + anodization.
5. The method of claim 1, wherein the step of acquiring a dataset comprises:
The method comprises the steps of establishing a geometric model of metal lap joint by adopting Abaqus software, wherein the geometric model consists of two layers of aluminum plates and a glue layer clamped between the two aluminum plates, and one layer of aluminum plate is partially lapped on the other layer of aluminum plate;
setting cohesive force contact between the aluminum plate of the geometric model and the adhesive layer, and carrying out grid division and boundary condition setting to construct a finite element model for metal-metal bonding;
verifying the accuracy of the finite element model to obtain a target finite element model with an error smaller than a preset error;
determining N simulation schemes according to multiple factors and multiple levels affecting metal-metal bonding performance, wherein the factor levels comprise overlap joint length, adhesive layer thickness and surface treatment mode, and N is an integer of more than or equal to 500 and less than or equal to 1000;
taking the N simulation schemes as input, and performing simulation by using the target finite element model to obtain the maximum tensile shear load corresponding to each simulation scheme;
based on the maximum tensile shear load corresponding to each simulation scheme, acquiring tensile shear strength corresponding to each simulation scheme;
and combining each simulation scheme with the pull-shear strength corresponding to each simulation scheme to obtain N simulation data, wherein each simulation data corresponds to one data packet, the simulation scheme is X data, and the pull-shear strength is Y data.
6. The method of claim 5, wherein the step of verifying the accuracy of the finite element model comprises:
step A, establishing an orthogonal table according to multi-factor multi-level affecting metal-metal bonding performance, wherein the multi-factor comprises overlap joint length, adhesive layer thickness and surface treatment mode;
step B, performing metal-metal bonding experiments according to M bonding schemes determined by the orthogonal table, and obtaining a maximum experimental tensile shear load corresponding to each bonding scheme, wherein M is an integer which is more than 8 and less than 15;
step C, taking the M cementing schemes as input, and performing simulation by using the finite element model to obtain the maximum simulation tensile shear load corresponding to each cementing scheme;
step D, calculating an error based on the maximum experimental pull shear load and the maximum simulation pull shear load corresponding to each cementing scheme;
and E, comparing the absolute value of the error with a preset error, and if the absolute value of the error is smaller than the preset error, taking the finite element model as a target finite element model.
7. The method of claim 6, wherein the step of verifying the accuracy of the finite element model further comprises:
And (C) when the absolute value of the error is larger than a preset error, updating the input parameters including the initial rigidity, critical traction and fracture energy of the corresponding adhesive layer when the cohesive force contact is set to obtain a new finite element model of metal-metal bonding, and repeating the steps (C) to (E).
8. The method of multi-factor analysis affecting metal-metal bonding properties according to any one of claims 1 to 7, wherein step (4) comprises:
establishing a one-to-one correspondence between root mean square errors and regression coefficients in each group of training results obtained based on multiple iterative training;
sequencing the root mean square errors in all the training results, and determining the minimum root mean square error;
and when the regression coefficient corresponding to the minimum root mean square error is larger than or equal to a preset regression coefficient, determining the Xgboost algorithm model corresponding to the minimum root mean square error as an optimal model.
9. The method according to claim 1, wherein the iteration termination condition of the multiple iteration training in the step (3) is that the number of iterations is 80 to 150.
10. A multi-factor analysis device for influencing metal-to-metal bonding performance, the analysis device comprising:
The acquisition module is used for acquiring a data set, wherein the data set comprises a plurality of data packets, each data packet comprises X data consisting of a multi-factor level affecting metal-metal bonding performance and Y data consisting of a tensile shear strength true value, and the X data and the Y data in each data packet have a one-to-one correspondence;
the construction module is used for constructing an Xgboost algorithm model and defining a hyper-parameter range of the Xgboost algorithm model;
the training module is used for carrying out iterative training for a plurality of times based on the data set and the Xgboost algorithm model to obtain a plurality of groups of training results for judging the reliability of the Xgboost algorithm model, wherein the training results comprise root mean square errors and regression coefficients, and the hyper-parameters of the Xgboost algorithm model corresponding to each iterative training are optimally updated in the hyper-parameter range of the set Xgboost algorithm model; wherein,,
each iterative training includes: cutting the data set into a training set and a testing set according to a preset proportion, and training the Xgboost algorithm model by taking the training set as input data to obtain a trained Xgboost algorithm model; inputting the test set as input data into a trained Xgboost algorithm model, and acquiring a tensile and shear strength predicted value corresponding to the X data of each data packet in the test set based on the X data of all the data packets in the test set; based on the pull-shear strength real value and the pull-shear strength preset value corresponding to the X data of each data packet in the test set, acquiring a training result corresponding to the iterative training, wherein the pull-shear strength real value is Y data corresponding to the X data;
The determining module is used for determining an optimal model based on a plurality of groups of training results obtained by repeated iterative training, wherein the root mean square error in the training results corresponding to the optimal model is minimum, and the regression coefficient is larger than or equal to a preset regression coefficient;
and the output module is used for leading a model interpretable tool SHAP into the optimal model and outputting interpretation results and prediction results, wherein the interpretation results comprise influence values of multiple factors affecting the metal-metal bonding performance on tensile shear strength respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310709558.6A CN116757031B (en) | 2023-06-15 | 2023-06-15 | Multi-factor analysis method and device for influencing metal-metal bonding performance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310709558.6A CN116757031B (en) | 2023-06-15 | 2023-06-15 | Multi-factor analysis method and device for influencing metal-metal bonding performance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116757031A true CN116757031A (en) | 2023-09-15 |
| CN116757031B CN116757031B (en) | 2024-02-09 |
Family
ID=87956628
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310709558.6A Active CN116757031B (en) | 2023-06-15 | 2023-06-15 | Multi-factor analysis method and device for influencing metal-metal bonding performance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116757031B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454673A (en) * | 2023-12-22 | 2024-01-26 | 中南大学 | Method and device for constructing metal material constitutive model by considering cooling rate |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2020101854A4 (en) * | 2020-08-17 | 2020-09-24 | China Communications Construction Co., Ltd. | A method for predicting concrete durability based on data mining and artificial intelligence algorithm |
| US20220004900A1 (en) * | 2018-11-26 | 2022-01-06 | Agency For Science, Technology And Research | Method and system for predicting performance in electronic design based on machine learning |
| US20220261520A1 (en) * | 2019-07-22 | 2022-08-18 | Jfe Steel Corporation | Quality prediction model generation method, quality prediction model, quality prediction method, metal material manufacturing method, quality prediction model generation device, and quality prediction device |
| CN115563824A (en) * | 2022-09-22 | 2023-01-03 | 吉林大学 | Method for predicting performance of dual-phase material composite tube based on machine learning |
| CN115599653A (en) * | 2022-10-28 | 2023-01-13 | 蕴硕物联技术（上海）有限公司（Cn） | Method, system and medium for constructing welding quality diagnosis model |
| CN115620838A (en) * | 2022-09-08 | 2023-01-17 | 广州大学 | Method for obtaining machine learning model combination for screening metal organic framework |
| CN116205310A (en) * | 2023-02-14 | 2023-06-02 | 中国水利水电科学研究院 | Soil water content influence factor sensitive interval judging method based on interpretable integrated learning model |
-
2023
- 2023-06-15 CN CN202310709558.6A patent/CN116757031B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220004900A1 (en) * | 2018-11-26 | 2022-01-06 | Agency For Science, Technology And Research | Method and system for predicting performance in electronic design based on machine learning |
| US20220261520A1 (en) * | 2019-07-22 | 2022-08-18 | Jfe Steel Corporation | Quality prediction model generation method, quality prediction model, quality prediction method, metal material manufacturing method, quality prediction model generation device, and quality prediction device |
| AU2020101854A4 (en) * | 2020-08-17 | 2020-09-24 | China Communications Construction Co., Ltd. | A method for predicting concrete durability based on data mining and artificial intelligence algorithm |
| CN115620838A (en) * | 2022-09-08 | 2023-01-17 | 广州大学 | Method for obtaining machine learning model combination for screening metal organic framework |
| CN115563824A (en) * | 2022-09-22 | 2023-01-03 | 吉林大学 | Method for predicting performance of dual-phase material composite tube based on machine learning |
| CN115599653A (en) * | 2022-10-28 | 2023-01-13 | 蕴硕物联技术（上海）有限公司（Cn） | Method, system and medium for constructing welding quality diagnosis model |
| CN116205310A (en) * | 2023-02-14 | 2023-06-02 | 中国水利水电科学研究院 | Soil water content influence factor sensitive interval judging method based on interpretable integrated learning model |
Non-Patent Citations (2)
| Title |
|---|
| ZHAO, Y.H: "Current research status of interface of ceramic-metal laminated composite material for armor protection", 《ACTA. METALL. SIN》 * |
| 刘春辉等: "金属变形过程的微观组织表征和多尺度建模方法", 《塑性工程学报》, vol. 28, no. 5 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454673A (en) * | 2023-12-22 | 2024-01-26 | 中南大学 | Method and device for constructing metal material constitutive model by considering cooling rate |
| CN117454673B (en) * | 2023-12-22 | 2024-03-12 | 中南大学 | Method and device for constructing metal material constitutive model by considering cooling rate |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116757031B (en) | 2024-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116757031B (en) | Multi-factor analysis method and device for influencing metal-metal bonding performance | |
| CN108645704B (en) | Method for calibrating microstructure parameters of metal material based on nanoindentation and finite element simulation inversion | |
| CN109388878A (en) | A kind of Prediction method for fatigue life comprehensively considering shot-peening effect | |
| CN106354898B (en) | A kind of weld fatigue Life Calculating Methods based on total strain energy metric density | |
| CN104809311A (en) | Structural part remaining life predicting method based on multi-factor fusion correction | |
| CN1890574A (en) | Apparatus and method for estimating state of charge of battery using neural network | |
| CN103970965A (en) | Test run method for accelerated life test of gas turbine engine | |
| CN105260574A (en) | Critical plane method fatigue failure criterion-based high-cycle multi-axial fatigue life prediction method | |
| CN106568647B (en) | A kind of Strength Forecast of Concrete method neural network based | |
| CN106327028A (en) | Terminal energy consumption prediction method and device | |
| CN116050623B (en) | Method for calculating and evaluating gas supply capacity of tight gas reservoir | |
| CN112051506B (en) | Similar product transferable sample screening method, system and application | |
| CN109858152B (en) | Method for predicting residual fatigue life of composite material containing initial delamination damage | |
| CN113591420B (en) | Simulation method and processor for true triaxial hydraulic fracturing experiment | |
| CN113255104B (en) | Efficient updating method for settlement prediction model | |
| CN105158147A (en) | Apparatus and method for testing ageing of material used for sealing ring | |
| Liu et al. | Prediction of shape change of semi-elliptical surface crack by fatigue crack growth circles parameter | |
| CN107784167B (en) | Multi-objective optimization method of integrated phenomenological structure based on dependency | |
| CN113094816B (en) | Method for constructing comprehensive working condition vibration spectrum and long-life test spectrum of armored vehicle | |
| CN115618676A (en) | Low-cycle fatigue crack simulation method and system based on continuous accumulated damage | |
| CN115391947A (en) | Virtual analysis method and system for composite material structure of rail transit vehicle | |
| CN103399974A (en) | Method for quantitatively comparing random vibration simulation data with experimental data | |
| Le et al. | Determination of a shape and size independent material modulus for honeycomb structures in additive manufacturing | |
| Strand et al. | Fourth-order analysis of a diffusive lattice Boltzmann method for barrier coatings | |
| CN110455902A (en) | A kind of method that the more standard specimens of environment measuring are quickly calibrated |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |